【摘要】 极化合成孔径雷达具备全天候的积雪观测能力,而且能提供丰富的极化特征用于积雪识别.选取2014年3月19日新疆玛纳斯河流域典型区Radarsat-2数据,首先对全极化SAR(Synthetic Aperture Radar)数据进行目标分解提取积雪极化特征,再利用J-M距离(Jeffreys-Matusita)进行特征选择,分析不同极化特征对积雪的可分性,最后利用最优特征集和支持向量机(Support vector machine,SVM)进行积雪识别.结果表明:Yamaguchi分解和Freeman分解的体散射分量、相干矩阵特征值和香农熵四种极化特征对积雪有较强的识别能力;多种极化特征联合识别相对于单一特征识别积雪具有较大优势,基于四种极化特征的积雪识别精度达到84%.利用极化特征进行积雪识别可获得较好效果,能够弥补可见光遥感难以识别云下积雪的不足.更多还原

【Abstract】 Polarimetric synthetic aperture radar(SAR)sensors can not only provide an all-weather snow observational capacity,but also provide a wealth of polarization characteristics,which have the potential to discriminate the snowcover from other natural scatters.In this paper,the data we acquired was Radarsat-2image in typical area of Manasi River Basin,Xinjiang Province on 19 March 2014.At first,we used polarimetric decomposition methods to extract polarimetric features for snow recognition.Secondly,Jeffreys-Matusita(J-M)distance was applied for feature selection.We analyzed the separability of different polarimetric features to discriminate between snow and snow-free are-as.At last,snow recognition was completed by using the best features and support vector machine(SVM).The results show that the volume scattering component of Yamaguchi and Freeman decomposition,eigenvalue of coherent matrix and Shannon entropy have strong recognition ability for snow,and compared with the single feature,combining several polarimetric features for snow recognition can obtain a better result and the accuracy based on the four polarimetric characteristics reached 84%.The snow identification by polarization features can acquirebetter effect and can remedy the limitationinsnow identification by visible spectral remote sensing under the cloud condition.更多还原